// ExactMatchNaive is a basic string searching algorithm that handles case
// sensitivity. Although naive, it still performs better than the combination
// of strings.ToLower + strings.Index for typical fzf use cases where input
// strings and patterns are not very long.
//
// We might try to implement better algorithms in the future:
// http://en.wikipedia.org/wiki/String_searching_algorithm
func ExactMatchNaive(caseSensitive bool, runes []rune, pattern []rune) (int, int) {
if len(pattern) == 0 {
return 0, 0
}
numRunes := len(runes)
plen := len(pattern)
if numRunes < plen {
return -1, -1
}
pidx := 0
for index := 0; index < numRunes; index++ {
char := runes[index]
if !caseSensitive {
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
if pattern[pidx] == char {
pidx++
if pidx == plen {
return index - plen + 1, index + 1
}
} else {
index -= pidx
pidx = 0
}
}
return -1, -1
}
// EqualMatch performs equal-match
func EqualMatch(caseSensitive bool, normalize bool, forward bool, text util.Chars, pattern []rune, withPos bool, slab *util.Slab) (Result, *[]int) {
lenPattern := len(pattern)
if text.Length() != lenPattern {
return Result{-1, -1, 0}, nil
}
match := true
if normalize {
runes := text.ToRunes()
for idx, pchar := range pattern {
char := runes[idx]
if !caseSensitive {
char = unicode.To(unicode.LowerCase, char)
}
if normalizeRune(pchar) != normalizeRune(char) {
match = false
break
}
}
} else {
runesStr := text.ToString()
if !caseSensitive {
runesStr = strings.ToLower(runesStr)
}
match = runesStr == string(pattern)
}
if match {
return Result{0, lenPattern, (scoreMatch+bonusBoundary)*lenPattern +
(bonusFirstCharMultiplier-1)*bonusBoundary}, nil
}
return Result{-1, -1, 0}, nil
}
// Implement the same sorting criteria as V2
func calculateScore(caseSensitive bool, normalize bool, text util.Chars, pattern []rune, sidx int, eidx int, withPos bool) (int, *[]int) {
pidx, score, inGap, consecutive, firstBonus := 0, 0, false, 0, int16(0)
pos := posArray(withPos, len(pattern))
prevClass := charNonWord
if sidx > 0 {
prevClass = charClassOf(text.Get(sidx - 1))
}
for idx := sidx; idx < eidx; idx++ {
char := text.Get(idx)
class := charClassOf(char)
if !caseSensitive {
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
// pattern is already normalized
if normalize {
char = normalizeRune(char)
}
if char == pattern[pidx] {
if withPos {
*pos = append(*pos, idx)
}
score += scoreMatch
bonus := bonusFor(prevClass, class)
if consecutive == 0 {
firstBonus = bonus
} else {
// Break consecutive chunk
if bonus == bonusBoundary {
firstBonus = bonus
}
bonus = util.Max16(util.Max16(bonus, firstBonus), bonusConsecutive)
}
if pidx == 0 {
score += int(bonus * bonusFirstCharMultiplier)
} else {
score += int(bonus)
}
inGap = false
consecutive++
pidx++
} else {
if inGap {
score += scoreGapExtention
} else {
score += scoreGapStart
}
inGap = true
consecutive = 0
firstBonus = 0
}
prevClass = class
}
return score, pos
}
func Spell(w string) {
last := len(w) - 1
for i, c := range w {
c = unicode.To(unicode.LowerCase, c)
if val, ok := Table[c]; ok {
fmt.Print(val)
} else {
fmt.Print(string(c))
}
if i != last {
fmt.Print(" ")
}
}
}
func normalizeReference(s string) string {
var buf bytes.Buffer
lastSpace := false
for _, r := range s {
if unicode.IsSpace(r) {
if !lastSpace {
buf.WriteByte(' ')
lastSpace = true
}
continue
}
buf.WriteRune(unicode.To(unicode.LowerCase, r))
lastSpace = false
}
return string(bytes.TrimSpace(buf.Bytes()))
}
// ExactMatchNaive is a basic string searching algorithm that handles case
// sensitivity. Although naive, it still performs better than the combination
// of strings.ToLower + strings.Index for typical fzf use cases where input
// strings and patterns are not very long.
//
// We might try to implement better algorithms in the future:
// http://en.wikipedia.org/wiki/String_searching_algorithm
func ExactMatchNaive(caseSensitive bool, forward bool, runes []rune, pattern []rune) Result {
if len(pattern) == 0 {
return Result{0, 0, 0}
}
lenRunes := len(runes)
lenPattern := len(pattern)
if lenRunes < lenPattern {
return Result{-1, -1, 0}
}
pidx := 0
for index := 0; index < lenRunes; index++ {
char := runeAt(runes, index, lenRunes, forward)
if !caseSensitive {
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
pchar := runeAt(pattern, pidx, lenPattern, forward)
if pchar == char {
pidx++
if pidx == lenPattern {
var sidx, eidx int
if forward {
sidx = index - lenPattern + 1
eidx = index + 1
} else {
sidx = lenRunes - (index + 1)
eidx = lenRunes - (index - lenPattern + 1)
}
return Result{int32(sidx), int32(eidx),
evaluateBonus(caseSensitive, runes, pattern, sidx, eidx)}
}
} else {
index -= pidx
pidx = 0
}
}
return Result{-1, -1, 0}
}
// ExactMatchNaive is a basic string searching algorithm that handles case
// sensitivity. Although naive, it still performs better than the combination
// of strings.ToLower + strings.Index for typical fzf use cases where input
// strings and patterns are not very long.
//
// We might try to implement better algorithms in the future:
// http://en.wikipedia.org/wiki/String_searching_algorithm
func ExactMatchNaive(caseSensitive bool, forward bool, runes []rune, pattern []rune) (int, int) {
if len(pattern) == 0 {
return 0, 0
}
lenRunes := len(runes)
lenPattern := len(pattern)
if lenRunes < lenPattern {
return -1, -1
}
pidx := 0
for index := 0; index < lenRunes; index++ {
char := runeAt(runes, index, lenRunes, forward)
if !caseSensitive {
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
pchar := runeAt(pattern, pidx, lenPattern, forward)
if pchar == char {
pidx++
if pidx == lenPattern {
if forward {
return index - lenPattern + 1, index + 1
}
return lenRunes - (index + 1), lenRunes - (index - lenPattern + 1)
}
} else {
index -= pidx
pidx = 0
}
}
return -1, -1
}
func ExampleTo() {
const lcG = 'g'
fmt.Printf("%#U\n", unicode.To(unicode.UpperCase, lcG))
fmt.Printf("%#U\n", unicode.To(unicode.LowerCase, lcG))
fmt.Printf("%#U\n", unicode.To(unicode.TitleCase, lcG))
const ucG = 'G'
fmt.Printf("%#U\n", unicode.To(unicode.UpperCase, ucG))
fmt.Printf("%#U\n", unicode.To(unicode.LowerCase, ucG))
fmt.Printf("%#U\n", unicode.To(unicode.TitleCase, ucG))
// Output:
// U+0047 'G'
// U+0067 'g'
// U+0047 'G'
// U+0047 'G'
// U+0067 'g'
// U+0047 'G'
}
func evaluateBonus(caseSensitive bool, runes []rune, pattern []rune, sidx int, eidx int) int32 {
var bonus int32
pidx := 0
lenPattern := len(pattern)
consecutive := false
prevClass := charNonWord
for index := 0; index < eidx; index++ {
char := runes[index]
var class charClass
if unicode.IsLower(char) {
class = charLower
} else if unicode.IsUpper(char) {
class = charUpper
} else if unicode.IsLetter(char) {
class = charLetter
} else if unicode.IsNumber(char) {
class = charNumber
} else {
class = charNonWord
}
var point int32
if prevClass == charNonWord && class != charNonWord {
// Word boundary
point = 2
} else if prevClass == charLower && class == charUpper ||
prevClass != charNumber && class == charNumber {
// camelCase letter123
point = 1
}
prevClass = class
if index >= sidx {
if !caseSensitive {
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
pchar := pattern[pidx]
if pchar == char {
// Boost bonus for the first character in the pattern
if pidx == 0 {
point *= 2
}
// Bonus to consecutive matching chars
if consecutive {
point++
}
bonus += point
if pidx++; pidx == lenPattern {
break
}
consecutive = true
} else {
consecutive = false
}
}
}
return bonus
}
// FuzzyMatch performs fuzzy-match
func FuzzyMatch(caseSensitive bool, forward bool, runes []rune, pattern []rune) Result {
if len(pattern) == 0 {
return Result{0, 0, 0}
}
// 0. (FIXME) How to find the shortest match?
// a_____b__c__abc
// ^^^^^^^^^^ ^^^
// 1. forward scan (abc)
// *-----*-----*>
// a_____b___abc__
// 2. reverse scan (cba)
// a_____b___abc__
// <***
pidx := 0
sidx := -1
eidx := -1
lenRunes := len(runes)
lenPattern := len(pattern)
for index := range runes {
char := runeAt(runes, index, lenRunes, forward)
// This is considerably faster than blindly applying strings.ToLower to the
// whole string
if !caseSensitive {
// Partially inlining `unicode.ToLower`. Ugly, but makes a noticeable
// difference in CPU cost. (Measured on Go 1.4.1. Also note that the Go
// compiler as of now does not inline non-leaf functions.)
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
pchar := runeAt(pattern, pidx, lenPattern, forward)
if char == pchar {
if sidx < 0 {
sidx = index
}
if pidx++; pidx == lenPattern {
eidx = index + 1
break
}
}
}
if sidx >= 0 && eidx >= 0 {
pidx--
for index := eidx - 1; index >= sidx; index-- {
char := runeAt(runes, index, lenRunes, forward)
if !caseSensitive {
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
pchar := runeAt(pattern, pidx, lenPattern, forward)
if char == pchar {
if pidx--; pidx < 0 {
sidx = index
break
}
}
}
// Calculate the bonus. This can't be done at the same time as the
// pattern scan above because 'forward' may be false.
if !forward {
sidx, eidx = lenRunes-eidx, lenRunes-sidx
}
return Result{int32(sidx), int32(eidx),
evaluateBonus(caseSensitive, runes, pattern, sidx, eidx)}
}
return Result{-1, -1, 0}
}
// FuzzyMatch performs fuzzy-match
func FuzzyMatch(caseSensitive bool, runes []rune, pattern []rune) (int, int) {
if len(pattern) == 0 {
return 0, 0
}
// 0. (FIXME) How to find the shortest match?
// a_____b__c__abc
// ^^^^^^^^^^ ^^^
// 1. forward scan (abc)
// *-----*-----*>
// a_____b___abc__
// 2. reverse scan (cba)
// a_____b___abc__
// <***
pidx := 0
sidx := -1
eidx := -1
for index, char := range runes {
// This is considerably faster than blindly applying strings.ToLower to the
// whole string
if !caseSensitive {
// Partially inlining `unicode.ToLower`. Ugly, but makes a noticeable
// difference in CPU cost. (Measured on Go 1.4.1. Also note that the Go
// compiler as of now does not inline non-leaf functions.)
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
if char == pattern[pidx] {
if sidx < 0 {
sidx = index
}
if pidx++; pidx == len(pattern) {
eidx = index + 1
break
}
}
}
if sidx >= 0 && eidx >= 0 {
pidx--
for index := eidx - 1; index >= sidx; index-- {
char := runes[index]
if !caseSensitive {
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
if char == pattern[pidx] {
if pidx--; pidx < 0 {
sidx = index
break
}
}
}
return sidx, eidx
}
return -1, -1
}
// ExactMatchNaive is a basic string searching algorithm that handles case
// sensitivity. Although naive, it still performs better than the combination
// of strings.ToLower + strings.Index for typical fzf use cases where input
// strings and patterns are not very long.
//
// Since 0.15.0, this function searches for the match with the highest
// bonus point, instead of stopping immediately after finding the first match.
// The solution is much cheaper since there is only one possible alignment of
// the pattern.
func ExactMatchNaive(caseSensitive bool, normalize bool, forward bool, text util.Chars, pattern []rune, withPos bool, slab *util.Slab) (Result, *[]int) {
if len(pattern) == 0 {
return Result{0, 0, 0}, nil
}
lenRunes := text.Length()
lenPattern := len(pattern)
if lenRunes < lenPattern {
return Result{-1, -1, 0}, nil
}
// For simplicity, only look at the bonus at the first character position
pidx := 0
bestPos, bonus, bestBonus := -1, int16(0), int16(-1)
for index := 0; index < lenRunes; index++ {
index_ := indexAt(index, lenRunes, forward)
char := text.Get(index_)
if !caseSensitive {
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
if normalize {
char = normalizeRune(char)
}
pidx_ := indexAt(pidx, lenPattern, forward)
pchar := pattern[pidx_]
if pchar == char {
if pidx_ == 0 {
bonus = bonusAt(text, index_)
}
pidx++
if pidx == lenPattern {
if bonus > bestBonus {
bestPos, bestBonus = index, bonus
}
if bonus == bonusBoundary {
break
}
index -= pidx - 1
pidx, bonus = 0, 0
}
} else {
index -= pidx
pidx, bonus = 0, 0
}
}
if bestPos >= 0 {
var sidx, eidx int
if forward {
sidx = bestPos - lenPattern + 1
eidx = bestPos + 1
} else {
sidx = lenRunes - (bestPos + 1)
eidx = lenRunes - (bestPos - lenPattern + 1)
}
score, _ := calculateScore(caseSensitive, normalize, text, pattern, sidx, eidx, false)
return Result{sidx, eidx, score}, nil
}
return Result{-1, -1, 0}, nil
}
// FuzzyMatchV1 performs fuzzy-match
func FuzzyMatchV1(caseSensitive bool, normalize bool, forward bool, text util.Chars, pattern []rune, withPos bool, slab *util.Slab) (Result, *[]int) {
if len(pattern) == 0 {
return Result{0, 0, 0}, nil
}
pidx := 0
sidx := -1
eidx := -1
lenRunes := text.Length()
lenPattern := len(pattern)
for index := 0; index < lenRunes; index++ {
char := text.Get(indexAt(index, lenRunes, forward))
// This is considerably faster than blindly applying strings.ToLower to the
// whole string
if !caseSensitive {
// Partially inlining `unicode.ToLower`. Ugly, but makes a noticeable
// difference in CPU cost. (Measured on Go 1.4.1. Also note that the Go
// compiler as of now does not inline non-leaf functions.)
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
if normalize {
char = normalizeRune(char)
}
pchar := pattern[indexAt(pidx, lenPattern, forward)]
if char == pchar {
if sidx < 0 {
sidx = index
}
if pidx++; pidx == lenPattern {
eidx = index + 1
break
}
}
}
if sidx >= 0 && eidx >= 0 {
pidx--
for index := eidx - 1; index >= sidx; index-- {
tidx := indexAt(index, lenRunes, forward)
char := text.Get(tidx)
if !caseSensitive {
if char >= 'A' && char <= 'Z' {
char += 32
} else if char > unicode.MaxASCII {
char = unicode.To(unicode.LowerCase, char)
}
}
pidx_ := indexAt(pidx, lenPattern, forward)
pchar := pattern[pidx_]
if char == pchar {
if pidx--; pidx < 0 {
sidx = index
break
}
}
}
if !forward {
sidx, eidx = lenRunes-eidx, lenRunes-sidx
}
score, pos := calculateScore(caseSensitive, normalize, text, pattern, sidx, eidx, withPos)
return Result{sidx, eidx, score}, pos
}
return Result{-1, -1, 0}, nil
}
func FuzzyMatchV2(caseSensitive bool, normalize bool, forward bool, input util.Chars, pattern []rune, withPos bool, slab *util.Slab) (Result, *[]int) {
// Assume that pattern is given in lowercase if case-insensitive.
// First check if there's a match and calculate bonus for each position.
// If the input string is too long, consider finding the matching chars in
// this phase as well (non-optimal alignment).
N := input.Length()
M := len(pattern)
switch M {
case 0:
return Result{0, 0, 0}, posArray(withPos, M)
case 1:
return ExactMatchNaive(caseSensitive, normalize, forward, input, pattern[0:1], withPos, slab)
}
// Since O(nm) algorithm can be prohibitively expensive for large input,
// we fall back to the greedy algorithm.
if slab != nil && N*M > cap(slab.I16) {
return FuzzyMatchV1(caseSensitive, normalize, forward, input, pattern, withPos, slab)
}
// Reuse pre-allocated integer slice to avoid unnecessary sweeping of garbages
offset16 := 0
offset32 := 0
// Bonus point for each position
offset16, B := alloc16(offset16, slab, N, false)
// The first occurrence of each character in the pattern
offset32, F := alloc32(offset32, slab, M, false)
// Rune array
offset32, T := alloc32(offset32, slab, N, false)
// Phase 1. Check if there's a match and calculate bonus for each point
pidx, lastIdx, prevClass := 0, 0, charNonWord
for idx := 0; idx < N; idx++ {
char := input.Get(idx)
var class charClass
if char <= unicode.MaxASCII {
class = charClassOfAscii(char)
} else {
class = charClassOfNonAscii(char)
}
if !caseSensitive && class == charUpper {
if char <= unicode.MaxASCII {
char += 32
} else {
char = unicode.To(unicode.LowerCase, char)
}
}
if normalize {
char = normalizeRune(char)
}
T[idx] = char
B[idx] = bonusFor(prevClass, class)
prevClass = class
if pidx < M {
if char == pattern[pidx] {
lastIdx = idx
F[pidx] = int32(idx)
pidx++
}
} else {
if char == pattern[M-1] {
lastIdx = idx
}
}
}
if pidx != M {
return Result{-1, -1, 0}, nil
}
// Phase 2. Fill in score matrix (H)
// Unlike the original algorithm, we do not allow omission.
width := lastIdx - int(F[0]) + 1
offset16, H := alloc16(offset16, slab, width*M, false)
// Possible length of consecutive chunk at each position.
offset16, C := alloc16(offset16, slab, width*M, false)
maxScore, maxScorePos := int16(0), 0
for i := 0; i < M; i++ {
I := i * width
inGap := false
for j := int(F[i]); j <= lastIdx; j++ {
j0 := j - int(F[0])
var s1, s2, consecutive int16
if j > int(F[i]) {
if inGap {
s2 = H[I+j0-1] + scoreGapExtention
} else {
s2 = H[I+j0-1] + scoreGapStart
}
}
if pattern[i] == T[j] {
var diag int16
if i > 0 && j0 > 0 {
diag = H[I-width+j0-1]
}
s1 = diag + scoreMatch
b := B[j]
if i > 0 {
// j > 0 if i > 0
consecutive = C[I-width+j0-1] + 1
// Break consecutive chunk
if b == bonusBoundary {
consecutive = 1
} else if consecutive > 1 {
b = util.Max16(b, util.Max16(bonusConsecutive, B[j-int(consecutive)+1]))
}
} else {
consecutive = 1
b *= bonusFirstCharMultiplier
}
if s1+b < s2 {
s1 += B[j]
consecutive = 0
} else {
s1 += b
}
}
C[I+j0] = consecutive
inGap = s1 < s2
score := util.Max16(util.Max16(s1, s2), 0)
if i == M-1 && (forward && score > maxScore || !forward && score >= maxScore) {
maxScore, maxScorePos = score, j
}
H[I+j0] = score
}
if DEBUG {
if i == 0 {
fmt.Print(" ")
for j := int(F[i]); j <= lastIdx; j++ {
fmt.Printf(" " + string(input.Get(j)) + " ")
}
fmt.Println()
}
fmt.Print(string(pattern[i]) + " ")
for idx := int(F[0]); idx < int(F[i]); idx++ {
fmt.Print(" 0 ")
}
for idx := int(F[i]); idx <= lastIdx; idx++ {
fmt.Printf("%2d ", H[i*width+idx-int(F[0])])
}
fmt.Println()
fmt.Print(" ")
for idx, p := range C[I : I+width] {
if idx+int(F[0]) < int(F[i]) {
p = 0
}
fmt.Printf("%2d ", p)
}
fmt.Println()
}
}
// Phase 3. (Optional) Backtrace to find character positions
pos := posArray(withPos, M)
j := int(F[0])
if withPos {
i := M - 1
j = maxScorePos
preferMatch := true
for {
I := i * width
j0 := j - int(F[0])
s := H[I+j0]
var s1, s2 int16
if i > 0 && j >= int(F[i]) {
s1 = H[I-width+j0-1]
}
if j > int(F[i]) {
s2 = H[I+j0-1]
}
if s > s1 && (s > s2 || s == s2 && preferMatch) {
*pos = append(*pos, j)
if i == 0 {
break
}
i--
}
preferMatch = C[I+j0] > 1 || I+width+j0+1 < len(C) && C[I+width+j0+1] > 0
j--
}
}
// Start offset we return here is only relevant when begin tiebreak is used.
// However finding the accurate offset requires backtracking, and we don't
// want to pay extra cost for the option that has lost its importance.
return Result{j, maxScorePos + 1, int(maxScore)}, pos
}
